FFT unit tests fail

[s-mayani]

The FFT unit tests do not pass for all cases on Cuda builds.

• With Cuda 12.1, they pass when using 1 GPU, but when using 2 GPUs or more, the error is an order of magnitude higher than the tolerance. Increasing the tolerance is not a solution because when increasing the number of GPUs even more, the error keeps increasing in order of magnitude.
• With Cuda 12.8 (tested by @biddisco), the error in the FFT unit test is very big (order of 200) for some of the tests.

It seems like the FFT unit tests passing or not depend a lot on the Cuda version being used as well as the amount of GPUs we use to test.